基于无人机多光谱数据的夏玉米综合水分指标估算模型

Estimation Model of Comprehensive Moisture Index for Summer Maize Based on UAV Multispectral Data

摘要:

农田水分精准监测对农业节水与产量保障至关重要，但现有技术多聚焦土壤或叶片/植株含水率等单一指标，缺乏对土壤-植株水分协同机制的系统性表征。本研究以关中平原夏玉米为研究对象，基于地面采样数据整合多深度土壤含水率、叶片含水率、叶片相对含水率、植株含水率等7项指标，分别利用熵权法和主成分分析法构建综合水分指标（Comprehensive moisture index）CMI1和CMI2，以反映农田土壤-植株的综合水分状况；同时结合无人多光谱数据计算并筛选敏感植被指数，利用随机森林（Random forest, RF）、支持向量机（Support vector machine, SVM）等机器学习算法构建无人机数据驱动的综合水分指标模型。结果表明：CMI1与CMI2均能有效反映夏玉米农田土壤-植株的综合水分状态，但CMI2对土壤-植株水分耦合特征的表征精度在拔节期、吐丝期等多数生育期优于CMI1。植被指数与综合水分指标的响应关系随生育期动态变化，各生育期筛选出的最佳植被指数及其与综合水分指标间的相关性不同，在拔节期、吐丝期、籽粒建成期和乳熟期与CMI的相关系数最高值分别可达0.761、0.795、0.769和0.771。RF模型在建模集和验证集中结果均较稳定，估算精度优于其他模型，能稳健地实现对夏玉米综合水分指标估算。本研究通过综合水分指标与机器学习模型性能的双重对比，构建的“多指标融合-无人机遥感驱动-动态建模”体系，为田块尺度综合水情监测提供了精准化技术方案，可支撑智慧农业灌溉决策。

Abstract:

Accurate farmland moisture monitoring is vital for agricultural water conservation and yield protection. However, existing technologies mainly focus on single indicators like soil or leaf/plant water content, lacking a systematic characterization of soil-plant water collaborative mechanisms. Taking summer maize in the Guanzhong Plain as the research object, seven indicators were integrated, including multi-depth soil water content, leaf water content, and plant water content through ground sampling. Two comprehensive moisture indices, CMI1 (using the entropy weight method) and CMI2 (using principal component analysis), were constructed to reflect the overall soil-plant moisture status. Sensitive vegetation indices were calculated and screened based on UAV multispectral data, and machine learning algorithms such as random forest (RF) and support vector machine (SVM) were applied to develop data-driven models for moisture estimation. The results showed that both CMI1 and CMI2 effectively reflected the comprehensive moisture status of summer maize farmland soil-plant systems, while CMI2 showed better characterization accuracy of soil-plant water coupling features than CMI1 in most growth stages (e.g.,jointing and silking stages). The response relationships between vegetation indices and comprehensive moisture indices varied dynamically with growth stages, and the highest correlation coefficients between optimal vegetation indices and CMI reached 0.761, 0.795, 0.769, and 0.771 in the jointing, silking, grain-filling, and milky stages, respectively. The RF model exhibited more stable performance in both modeling and validation sets, with estimation accuracy superior to other models, enabling robust estimation of comprehensive moisture indices for summer maize. The research result presented a “multi-index integration-UAV remote sensing-dynamic modeling” framework through dual performance comparisons of moisture indices and machine learning models, offering precise field-scale monitoring solutions for smart irrigation decisions.